The source VM remains running while conversion occurs. This is far more complex. The converter installs an agent (or uses a hypervisor’s native API) to take a point-in-time snapshot. It then reads the snapshot’s blocks, converts them, and writes to the target. Meanwhile, it tracks all new writes to the source disk (the “dirty block log”). Once the initial copy is complete, the converter pauses the source VM briefly, syncs the dirty blocks, transfers control, and boots the target VM. VMware vCenter Converter’s “hot cloning” is a classic example. This minimizes downtime to seconds but risks data inconsistency if the dirty block tracking fails.
We are seeing the emergence of —where a management layer (e.g., Red Hat Virtualization, OpenStack) treats all VM formats as ephemeral. The user requests a VM; the orchestrator pulls the source from any format (VMDK, VHDX, raw) and converts it on-the-fly during streaming boot. This eliminates the distinction between “conversion” and “execution.” vmconverter
The first wave of cloud migration relied on “lift and shift”—taking on-premise VMs and converting them to cloud-native instances. AWS VM Import/Export, Azure Site Recovery, and Google Cloud Migrate all embed VMConverter logic. They convert VMDK/VHD to AMI or managed disk formats, reconfiguring the bootloader for cloud-init and replacing the kernel for cloud-optimized drivers. Without these converters, the hybrid cloud would be a patchwork of incompatible silos. The source VM remains running while conversion occurs